Chapter Preamble
3 Direct effects of an altered plant community with smooth brome invasion on soil bacterial community structure and
3.3 Materials and Methods .1 Field Site
3.3.4 Statistical analysis
The relationship between smooth brome and bacterial species richness and evenness was examined using linear mixed models. Smooth brome shoot biomass was used as an indicator of smooth brome abundance and was used along with horizon as explanatory variables. “Plot” was included as a random factor. Mixed models were run in R (R Core Development Team, 2012) using the nlme library (Pinheiro et al., 2012). Fitted vs. residual and qq plots were used to ensure appropriate model fit.
This initial assessment showed significant effects of smooth brome on bacterial community richness and evenness. To investigate the potential mechanisms underlying these effects, a multi‐group structural equation model (SEM) was fit using plant productivity and soil composition data as observed variables. SEM was chosen because it allowed for separation and testing of the direct and indirect relationships between intercorrelated variables (Grace, 2006;
Lamb et al., 2011a). The first step in SEM is to develop an initial path model based on prior theoretical knowledge about the system. The second step is to test for fit between the implied covariance structure of the theoretical model and the actual covariance structure of the data. Initial fit between the model and data provides strong support for the theoretical relationships being tested. A multi‐group model is appropriate for this dataset as we have two subsets of data (A and B horizon)
collected from the same sample points. In a multi‐group SEM, models are initially constrained so that path coefficients are equal between groups. These constraints can then be progressively released to improve model fit. A difference in path
coefficients between horizons indicates a significant difference between horizons in the biological process represented by that path.
The initial SEM model was developed to examine the influence of changing plant shoot community composition on soil bacterial community structure (Figure 3.2). Smooth brome shoot biomass and plant species richness were used as
indicators of plant community composition. We hypothesized that brome may influence plant richness directly, or indirectly through changes in litter biomass. As plant species richness may also be influenced by site productivity, we included a direct relationship between A horizon depth (an indicator of long‐term site
productivity) and native plant species richness. As different plant species produce litter of differing quality and composition (Cornelissen, 1996; Wardle, 2002) we included direct relationships from brome biomass and native species richness to litter C:N ratio, a measure of litter quality. As leaf litter and root decomposition are important sources of organic carbon and nitrogen, we included direct relationships from litter quality and quantity and root biomass to soil organic carbon and total nitrogen. Soil organic carbon and total nitrogen were used as predictors of bacterial community richness and evenness, as resource availability is known to influence bacterial community composition (Drenovsky et al., 2004; Fierer et al., 2003).
Changes in root biomass may also influence bacterial community structure as plant roots and soil bacteria are strongly linked (Wardle et al., 2004). We also included bivariate (non‐directed) relationships between soil organic carbon and total nitrogen. Mean values for all variables used in the models are given in Table 3.1.
Figure 3.2 Initial structural equation model. Single‐headed arrows represented directed relationships and double‐headed arrows represent bivariate relationships (undirected).
Table 3.1 Mean and standard deviation for all variables included in structural equation modeling analysis. For horizon‐level data, asterisks indicate differences in mean values between horizons (t‐test, p<0.001).
Variable Mean ± standard deviation Smooth brome shoot biomass
(g/m2) 265 ± 211.2
Plant richness 11 ± 3.55
A horizon depth (cm) 12 ± 4.16 Litter biomass (g/m2) 336 ± 160.0 Litter C:N ratio 23.6 ± 5.60 Root biomass (g/m2)
A Horizon B Horizon
1612 ± 749.6**
433 ± 302.6 Soil organic carbon (%)
A Horizon B Horizon
6.0 ± 1.90**
2.5 ± 0.787 Total soil nitrogen (%)
A Horizon B Horizon
0.61 ± 0.168**
0.27 ± 0.0922 Bacterial species richness
A Horizon B Horizon
616 ± 122.8**
508 ±118.6 Bacterial community evenness
A Horizon B Horizon
0.63 ± 0.0490**
0.59 ± 0.0534
Prior to fitting the SEM, relationships were checked for linearity using general linear models that included a quadratic term. Significant quadratic terms were found for relationships involving root biomass; these relationships were linearized by log transforming root biomass. Smooth brome shoot biomass, litter biomass, and bacterial species richness were divided by 1.0x104 to equalize variances. The SEM models were fit using the lavaan library in R (Rosseel et al., 2012). The SEM model was built step‐wise, first fitting a single SEM model with only bacterial species richness. As this model had adequate fit, it was fit as a multi‐group model with all parameters constrained to be equal. This model did not have
adequate fit (255=99.0, p<0.001), but through sequential release of parameter constraints with high standardized residuals, it reached adequate fit (532=62.2, p=0.180). A second model was fit with bacterial community evenness replacing richness. The initial single evenness model had adequate fit, but the multi‐group model did not initially have good fit (552=100.0, p<0.001). Through release of parameters, it reached adequate fit (532=62.2, p=0.180). Both the richness and evenness models showed that none of the variables predicting bacterial richness or evenness were important. To confirm that smooth brome was in fact influencing these variables, direct relationships were added from brome shoot biomass and plant richness to bacterial richness and evenness. These ad hoc pathways were added to represent an unknown mechanism rather than a direct theoretical relationship. Increased 2 values and decreased Akaike’s Information Criterion (AIC) values were used to determine if these added direct relationships improved model fit (Akaike, 1974).
The relationships between plant and bacterial community composition were examined using non‐metric multidimensional scaling (NMS). NMS was used as it is robust for ecological, non‐normal datasets (McCune and Grace, 2002). Plant
community data were ordinated using the Sorensen (Bray‐Curtis) distance metric in PC‐Ord 5 (Kruskal, 1964; Mather, 1976; McCune and Mefford, 2006). Separate ordinations were run for each horizon due to different missing sample points in the A and B horizon bacterial datasets. Ordinations were completed using random starting configurations and 50 runs with real data. For both horizons, a two
dimensional solution was chosen (Stress A Horizon=16.6, Stress B Horizon=15.9), and the Monte Carlo test was significant (p=0.0196). The final solution was based on 200 iterations. Both ordinations were rotated graphically so that Axis 1 was most
correlated with smooth brome abundance. To explore the relationships between the plant and bacterial community, a joint plot (r2>0.1) of bacterial abundance
aggregated at the phylum and order level was overlaid to examine broad
relationships between the plant and bacterial community. Some OTUs could not be classified to species or were classified to unnamed taxonomic groups, and therefore OTUs were named to known taxonomic level. OTUs that were classified to a
taxonomic group with certainty lower than 75% were changed to unclassified for that level of taxonomic resolution.
3.4 Results
3.4.1 Influence of smooth brome on bacterial richness and evenness
Bacterial species richness increased in both the A and B horizon with increasing smooth brome biomass (Figure 3.3). Species richness (R) was higher in